THE MOST DISTANT STARS IN THE MILKY WAY

The outer halo of our Milky Way (MW) has yet to be comprehensively mapped. Surveys such as the Sloan Digital Sky Survey (SDSS; York et al. 2000) have yielded large samples of metal-poor main sequence turnoff stars, which have been used to map out the MW's inner halo (e.g., Bell et al. 2008) to distances of d ≲ 50 kpc over large areas of sky. Other stellar tracers, such as RR Lyrae (RRL) stars (e.g., Drake et al. 2013), red giant branch stars (e.g., Helmi et al. 2003; Xue et al. 2014) and blue horizontal branch (BHB) stars (e.g., Schlaufman et al. 2009), have also been used to map the Galactic halo to distances out to d ∼ 120 kpc. However, very few stars at d ≳ 120 kpc have been identified mostly due to the faint limits of large surveys (r ≲ 21). Table 1 lists the known stars at d > 120 kpc.

This severely incomplete picture of the Galaxy's outermost reaches limits our understanding of halo formation and evolution. Stars are not expected to be able to form in situ at outer halo distances (e.g., Zolotov et al. 2009). Instead, lone stars found at truly large distances (d ≳ 200 kpc) could mark the presence of outer halo substructure from accreted dwarf galaxies (e.g., Bullock & Johnston 2005; Majewski et al. 2003), or result from a star being ejected from the Galactic center or disk (e.g., Brown et al. 2005). If they are bound to the MW, distant halo stars could be used to measure the MW's virial mass, which is uncertain by a factor of ∼3, ranging from ∼0.7 to 2 × 10¹² M_☉ (Sofue 2012; Deason et al. 2012b; Burch & Cowsik 2013; Irrgang et al. 2013).

M giants currently offer the best opportunity to probe the MW's outer halo with existing surveys. They are intrinsically bright, with typical bolometric luminosities of log L/L_☉ ∼ 3–4. Given their relatively cool effective temperatures, near infrared (NIR) surveys, such as the Two-Micron All Sky Survey (2MASS; Skrutskie et al. 2006) and the UKIRT Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007), are especially sensitive to M giants. M giants in 2MASS were used to extensively map out the Sagittarius dwarf galaxy (Sgr; Majewski et al. 2003; Law & Majewski 2010). SDSS observations of M giants were also used to map out the Sgr tidal remnants, including the structure at distances near ∼100 kpc (Yanny et al. 2009; Belokurov et al. 2014). The UKIDSS data, which has a faint limit 3–4 mag deeper than 2MASS, is sensitive to the most luminous M giants to distances beyond the MW's virial radius (∼180–300 kpc; Klypin et al. 2002; Deason et al. 2012a; Piffl et al. 2014, and references therein).

In Bochanski et al. (2014, hereafter Paper I), we assembled a catalog of 404 M giant candidates, which were selected based on their NIR colors from the UKIDSS Large Area Survey (ULAS), optical colors from SDSS, and proper motions. In this Letter, we report on the discovery of two extremely distant M giants from this catalog, ULAS J001535.72+015549.6 and ULAS J074417.48+253233.0 (hereafter ULAS J0015+01 and ULAS J0744+25). These stars were spectroscopically confirmed as M giants, at distances of ∼270 and 240 kpc (and at least 180 and 130 kpc, respectively, depending on metallicity). They are the first MW stars identified beyond 200 kpc, and potentially beyond the virial radius. We present our observations in Section 2. Our spectral type determinations, distance estimates and radial velocity measurements are described in Section 3. In Section 4 we discuss our hypotheses for the origins of these stars.

The details of our M giant candidate selection are contained in Paper I, but are briefly provided here. We selected M giants using NIR color cuts, similar to the Majewski et al. (2003) and Sharma et al. (2010) color cuts, but shifted to reduce contamination from K/M dwarfs and K giants. Our NIR targets were matched to SDSS observations, which were used to discriminate against quasars. Reddening is small throughout the sample. For example, E(B − V) values for ULAS J0015+01 and ULAS J0744+25 are 0.03 and 0.04, respectively. We astrometrically screened out M dwarfs using the SDSS-USNOB proper motion catalogs (Munn et al. 2004), supplemented by proper motions derived from SDSS-UKIDSS astrometry. After all cuts were applied, 404 M giant candidates remained. Our initial study netted five M giants, and our spectroscopic campaign is ongoing.

During 2014 November 12–14, we obtained spectra of 32 stars using the Red Channel Spectrograph (RCS; Schmidt et al. 1989) at the 6.5 m MMT observatory at Mount Hopkins, Arizona. We used the RCS in single order mode, employing the 10 slit, 1200 lines mm⁻¹ grating centered at 8400 Å and LP-530 long-pass filter, resulting in a resolution of R ∼ 5000. Due to variable and significant cloud coverage during the first night, we observed relatively bright (V < 10) stars previously classified as M giants and M dwarfs to aid in spectroscopic classification. These stars, which serve as high signal-to-noise (S/N) spectral standards, are listed in Table 2. Over the remaining 2 nights, we observed 15 M giant candidates, listed in Table 2. Science observations ranged from 180s to 7200s of total exposure time, broken into multiple exposures to aid in cosmic ray rejection. The data were reduced using an IDL software package based on the MASE (Bochanski et al. 2009) reduction pipeline. Each single-order observation was bias-corrected, flat-fielded, wavelength calibrated, and optimally extracted. Wavelength calibrations were obtained using the HeNeAr arc lamp, and corrected to the heliocentric rest frame. Flux calibration was computed by comparing to standards, with at least one standard being observed per night. The typical S/N of our science observations ranged from ∼10 to 50. The spectra of many of the stars obtained during our run are shown in Figure 1.

Zoom In Zoom Out Reset image size

3.1. Spectral Type Estimates

Spectral types for our science targets were estimated using two methods. First, each science spectrum was compared to the 17 giant and dwarf spectra. We visually compared each standard-science spectral pair, and computed the χ² residuals over the wavelength regime shown in Figure 1. The spectra of ULAS J0015+01 and ULAS J0744+25 agreed more closely with the giant standards, both visually and with respect to χ² residuals. M dwarf and M giant template spectra were also constructed by coadding individual giant and dwarf spectra. Each standard observation (gray lines in Figure 1) was normalized with a fourth-order polynomial to remove the stellar continuum prior to coaddition. The coadded template spectra are shown in Figure 1, along with the spectra of ULAS J0015+01 and ULAS J0744+25. The dominant atomic spectral features in this wavelength regime are the Na i doublet near 8200 Å and the Ca ii triplet near 8500 Å. TiO gives rise to the bandheads near 8400 Å, which are easily observed in the M giant spectra, as they change strength drastically throughout the sequence. Telluric features are notable as well. The absorption seen at 8227 Å is due to terrestrial water vapor (Albers 1974). The emission features near 8345, 8400, 8430, 8505, and 8780 Å are due telluric OH emission lines (Cosby et al. 2006).

In Figure 2, we compare the spectra of the distant M giants to the M giant (top) and M dwarf (bottom) templates. The M giant template is a better match to both stars. While the S/N in the ULAS J0015+01 spectrum is only ∼10, the Na i doublet near 8200 Å is not visible. The spectrum of ULAS J0744+25 has a higher S/N (∼20), and is well matched by the M giant template. While M dwarfs exhibit Ca ii absorption, the M giant template is a better match on the blue side of the spectrum, and displays no Na i absorption. Furthermore, if either star was an M dwarf, the Na i doublet would be of similar strength to the water vapor near 8227 Å. However, there are no strong absorption bands in this regime. We note that low-mass subdwarfs also have weakened Na i absorption. Recently, Savcheva et al. (2014) prepared a catalog of 3517 low-mass subdwarfs with SDSS spectra. Less than 1% of these stars would have passed our color and proper motion cuts. We visually inspected the SDSS spectra of subdwarfs that did, and all displayed prominent Na i absorption. Thus, ULAS J0015+01 and ULAS J0744+25 were classified as M giants. The M giant with the closest match to both stars, both visually and with respect to χ² was EH Cet, an M4 III giant with an M_J = −4.6 (van Leeuwen 2007).

Zoom In Zoom Out Reset image size

3.2. Distance Estimates

3.3. Radial Velocity Measurements

The origins of ULAS J0015+01 and ULAS J0744+25 are interesting, given their large distances from the MW. They are at least three times as distant as the Large Magellanic Cloud, and potentially more distant than Leo I and Leo II (Harrington & Wilson 1950). Both stars are near the MW's virial radius, where gas densities are extremely low, and star formation is virtually non-existent. Thus, it would be overwhelmingly unlikely that either star formed in situ.

Given the largely accepted accretion model for the formation of the MW's halo, the most natural hypothesis for the origin of these stars is accretion from a dwarf galaxy (e.g., Bullock & Johnston 2005; Zolotov et al. 2009). Such tidal stripping is well mapped in the inner halo, which is dominated by the Sgr dwarf and its well-mapped tidal tails (Majewski et al. 2003; Belokurov et al. 2014). However, only one stream has been identified at d > 100 kpc (Drake et al. 2013). We compared the position and radial velocities of both stars to a model of Sgr (Law & Majewski 2010). Unlike the closer M giants associated with Sgr in Paper I, we did not definitively associate these M giants with Sgr. However, we do note that both stars lie close to the Sgr plane, with ULAS J0744+25 at (Λ, B)_Sgr ∼ (190.2, 4.7), and ULAS J0015+01 at (Λ, B)_Sgr ∼ (81.8, −16.2). Thus, an association with Sgr cannot be ruled out. Furthermore, ULAS J0015+01 is spatially coincident with the apocenter of the Sgr trailing arm, while ULAS J0744+25 is spatially coincident with the Pisces overdensity (Sharma et al. 2010). Despite being within ∼10° of these structures, both stars have much larger distances.

Despite differences in formation history and halo masses, simulations predict that the majority of stars at d > 150 kpc have been accreted less than 9 Gyr ago, suggesting recent accretion is important in the outer halo (Zolotov et al. 2009). Models also predict significantly less substructure with d ≳ 160 kpc and a stellar density over four orders of magnitude smaller than the solar radius (Bullock & Johnston 2005). While the predicted amount of substructure in the outer halo is small, our experiment was designed to target any trace of this population. Therefore, it is possible that ULAS J0015+01 and ULAS J0744+25 have been stripped from substructure accreted in the last 10 Gyr. With more complete maps and UKIDSS sky coverage, this hypothesis can be more fully explored.

An alternative hypothesis for these stars' origin is ejection from the MW. Both star–binary (i.e., disk ejection) and star–black hole (i.e., hypervelocity) interactions can eject stars (i.e., Zhang et al. 2013; Perets & Šubr 2012). To test this mechanism, we calculated the time it took each star to travel to its current position and velocity, assuming it was formed near the solar circle and was ejected. Since the halo is the dominant potential at these distances, we used a spherical Navarro-Frenk-White halo (Navarro et al. 1996) to model the Galactic potential:

$$\begin{equation} \Phi (r) = -\frac{G M_{\mathrm{vir}}}{a f(c)} \frac{\ln (1+r/a)}{r/a}, \end{equation} \tag{ 1 }$$

where the scale radius a is related to the concentration c of the halo via a ≡ r_vir/c, and the concentration constant f(c) is

$$\begin{equation} f(c) \equiv \ln (1+c) - \frac{c}{1+c}. \end{equation} \tag{ 2 }$$

Recent estimates of the concentration parameter c, range from ∼18 to 24 (Battaglia et al. 2005; Deason et al. 2012b; Smith et al. 2007). As mentioned above, the current estimated for the virial mass of the MW range from ∼0.7 to 2 × 10¹² M_☉ (Sofue 2012; Deason et al. 2012b; Burch & Cowsik 2013; Irrgang et al. 2013). Thus, we adopted the middle values of the MW's mass (1.4 × 10¹² M_☉) and concentration parameter (c = 20) as fiducial values for the MW potential.

First, we calculated the speeds with which the stars would have left the disk, along with the corresponding apocenters. Assuming each star started near the solar circle (8 kpc), the initial Galactocentric velocities were similar: v_{r, J0015 + 01} = 572 km s⁻¹ and v_{r, J0744 + 25} = 563 km s⁻¹. The additional potential terms of the disk and bulge were ignored, but they would serve to increase these speeds. Even without these Galactic components, the necessary speed for each star to reach its current distance is comparable to the measured escape velocity of the MW at the solar circle (500–600 km s⁻¹; Smith et al. 2007). This suggests that the stars were unlikely to be ejected, as the mechanisms that can produce these speeds are relatively rare. Hypervelocity star ejection rates are ∼1× 10⁻⁴ yr⁻¹ (i.e., Zhang et al. 2013), while the rate of 600 km s⁻¹ disk runaway ejections is ∼1× 10⁻⁶ yr⁻¹ (i.e., Perets & Šubr 2012).

Next, we calculated the time needed to reach each stars' current position. Using the fiducial model, it would take both stars ∼1.5 Gyr to reach their current position, while moving outward. We varied the mass and concentration of the halo to compute a range of travel times, which were the same for both stars and varied from ∼1 to 3 Gyr. We will directly test the ejection hypothesis as our sample grows. If ejection is important in placing M giants at these distances, there should be more stars seen in the direction of rotation, since they would gain a ∼240 km s⁻¹ boost from the MW's rotation.

Another hypothesis for the origin of these stars is their membership in a previously unseen dwarf galaxy. Nearby MW dwarf spheroidals (Draco, Ursa Minor, Sculptor, Carina, Fornax) do not host obvious M giant populations as cataloged by 2MASS with our NIR cuts applied. This paucity of 2MASS M giants in MW dwarfs is not surprising, given 2MASS's faint limit and the low metallicities of the dwarfs. However, Carina (Fe/H =−1.72; McConnachie 2012) and Fornax (Fe/H = −0.99; McConnachie 2012) each have one to two candidate M giant stars in 2MASS at the correct photometric distances for membership. Assuming that a galaxy capable of hosting an M giant has a mean [Fe/H] ≳ −1.0, we used the stellar mass-metallicity relation to estimate a M_V ≲ −13 for any unseen dwarf galaxy that could host these stars (Kirby et al. 2013). For such a dwarf to have escaped detection, it would have to be extremely low surface brightness (μ_V, ≳ 30 mag arcsec⁻²). Although unseen, the existence of such ghostly objects has been predicted (Bullock et al. 2010; Bovill & Ricotti 2011). To test this hypothesis, we constructed the SDSS g, g − r color–magnitude diagram (CMD) within 15 arcmin of each star. These CMDs were binned and subtracted with a control CMD from a nearby field of the same area. Visual inspection of the subtracted CMDs revealed no obvious detection of any associated red giant branches or Sgr turnoff stars near either M giant. Multi-object spectroscopy of surrounding stars with r > 20 mag could provide a test this hypothesis, along with deep search for RR Lyrae stars near each M giant.

After observing 15 M giant candidates, our spectroscopic campaign confirmed four new M giants in the MW's halo. This selection efficiency is in-line with our initial study (∼20%, Paper I) and we expect that ∼70 M giants will be recovered from our sample. By identifying more of these stars spectroscopically, we hope to further refine our selection criteria. This will be important for identifying M giants in the next generation of large surveys, such as Gaia and LSST (Perryman et al. 2001; Ivezic et al. 2008). While these stars will be too faint for reliable parallax measurements with Gaia, they may be able to recover proper motions in the halo. For example, an M giant at Gaia's faint limit (G = 20), with a distance of 200 kpc and a tangential velocity of 100 km s⁻¹ will have a proper motion of ∼100 μas yr⁻¹. Gaia's expected proper motion uncertainty at G = 20 will be ∼50 μas yr⁻¹, resulting in a 2σ measurement (de Bruijne 2012).

J.J.B. and B.W. thank the NSF for support under grants NSF AST-1151462 and PHYS-1066293. A.A.W acknowledges NSF grants AST-1109273, AST-1255568, and the RCSA's Cottrell Scholarship. We thank Jonathan Hargis, Alis Deason, Wyn Evans, Vasily Belokurov and Kathyrn Johnston for helpful conversations.